This post about the antlr simple example shows how to create and us a grammar for java.
However, this intermixes the grammar and the Java source code in the Exp.g source.
My Question is, Is it possible to decouple the grammar file from the target language, so that the one grammar file can be used for generating multiple Java, Scala, C++, etc Lexers/Parsers?
It depends mostly on the reason why target code is used in the grammar. Is it only action code to do something with the found tokens (e.g. building a symbol table or alternative tree representation) then is indeed no problem do remove such native code and do the processing afterwards (using a parse tree walker or visitor).
However, predicates are a different. They are used to guide the parser and also require native code. What you can do is to move all the native code into a base class from which your generated parser derives. You then only need to re-write this base class in your target language and keep the grammar mostly free of native code (except for a single function call, which invokes the native code).
This approach has the advantage that no additional library reference is necessary (#include in C/C++, import in other languages), which also is native code preventing use for multiple targets.
